1 / 27

I2 and NDIIPP: Internet2 Infrastructure in Support of the National Preservation Agenda

I2 and NDIIPP: Internet2 Infrastructure in Support of the National Preservation Agenda. Andy Boyko – Library of Congress Jane Mandelbaum – Library of Congress Robert McDonald - SDSC David Minor - SDSC Emilio Valente - SDSC. Outline. October 11, 2007 8:45 am – 10:00 am

naeva
Download Presentation

I2 and NDIIPP: Internet2 Infrastructure in Support of the National Preservation Agenda

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. I2 and NDIIPP: Internet2 Infrastructure in Support of the National Preservation Agenda Andy Boyko – Library of Congress Jane Mandelbaum – Library of Congress Robert McDonald - SDSC David Minor - SDSC Emilio Valente - SDSC

  2. Outline October 11, 2007 8:45 am – 10:00 am • LC Background with Internet2 (Jane/Andy) – 10 mins • Pilot Data Center Project w/SDSC (David) – 10 mins • Network Optimization and Data Transfer (Andy/Emilio) – 25 mins • NDIIPP I2 and Future (Jane) – 10 mins

  3. LC and I2: How We Got Here • Early adopter of the Internet protocols and philosophies • Goal of “library without walls” • Long history of partner exchanges and educational outreach across the Internet • Early data transfers focused on bibliographic data • 22 million objects now online and growing • Creation of “universal digital library” will require even more content exchange over the Internet.

  4. LC Partners: NDIIPP • National Digital Information Infrastructure Preservation Program • NDIIPP: “Infrastructure” is the base • Content transfer is one of the partnership services. • How do we make that service a success for all our current and future partners? • Build on the experience with the LC-SDSC project.

  5. SDSC and I2 • One of original five NSF supercomputer centers (1985) • Supports High Performance Computing Systems • Supports Data Applications for Science, Engineering, Social Sciences, Cultural Heritage Institutions • 2+ PB Disk Capacity • 25+ PB Tape Capacity • Connections to: • I2 Abilene Network • NLANR • TeraGrid Network

  6. Data Center for Library of Congress Digital Holdings: A Pilot Project Library of Congress:Office of Strategic Initiatives(National Digital Information Infrastructure and Preservation Program) University of California, San Diego:San Diego Supercomputer Center and UCSD Libraries

  7. Project Overview:“Building Trust in a Third Party Data Repository” “… demonstrate the feasibility and performance of current approaches for a production digital Data Center to support the Library of Congress’ requirements.” - Pilot project to be completed in 1 year - $ 1 million - Transfer, store and study multiple TBs of data

  8. Data Collection: Prints and Photographs Division Prokudin-Gorskii Photographs http://www.loc.gov/exhibits/empire/

  9. Data Collection: Prints and Photographs Division Characteristics of the collection • Different file types based on the original pieces • Recreations of projections, based on files • File structure based on the collection In many ways, a good example of digital memory:extending the lifespan and accessibility of a traditionalcollection using digital mechanisms.

  10. Data Collection: Prints and Photographs Division What did we do with collection? • Replicated structure of filesystem in remote location • Provided a new front end • Provided extensive logging and monitoring • Tasks accomplished using SRB

  11. Data Collection:Web Archiving and Preservation Project Characteristics of the collection A living snapshot of this moment in history. These “documents” exist nowhere else. • 6TB of of “born digital” materials • Library had never indexed this much at once • Special file format and software installations

  12. Data Collection:Web Archiving and Preservation Project What did we do with collection? • Indexed all data by re-writing indexing software – took it from 30+ days of compute time to 7 days • Installed and configured Wayback web access to replicate their environment • Performed usability studies comparing our two sites.

  13. Content in Motion • Initial project plan specified disk-based data transfer from LC to SDSC • 6TB+, spread across dozens of hard disks • Copying, testing, packing, shipping, re-testing: time-consuming, potential for error/loss at all steps • When LC acquired Abilene connection, at time of disk transfer, expanded project scope to test and compare network transfers • Chose goal of at least 1TB/day rates (~100Mb/s or better) • Unit of transfer: packages of 300-600GB (corresponding to original hard disks)

  14. Data Transfer: Initial Network Environment

  15. Networking LC-SDSC: Initial Status • lctb1:/users/u3/evalente-4 >ping -s xxx.xxx.xxx.x (LC address) • PING xxx.xxx.xxx.x: 56 data bytes • 64 bytes from www.loc.gov (xxx.xxx.xxx.x): icmp_seq=0. time=76. ms • 64 bytes from www.loc.gov (xxx.xxx.xxx.x): icmp_seq=1. time=76. ms • 64 bytes from www.loc.gov (xxx.xxx.xxx.x): icmp_seq=2. time=76. ms • 64 bytes from www.loc.gov (xxx.xxx.xxx.x): icmp_seq=3. time=76. ms • Initial tests with HTTP protocol, downloading a 70 MBytes file. • Average speed was poor: 200 Kb/s. • Iperf Test: 3.32 Mbits/sec • Client connecting to 132.249.21.26, TCP port 6023 • TCP window size: 256 KByte (WARNING: requested 1.00 MByte) • ------------------------------------------------------------ • [ 3] local 192.168.1.43 port 55613 connected with 132.249.21.26 port 6023 • [ 3] 0.0-10.8 sec 4.25 MBytes 3.32 Mbits/sec

  16. Pathway Bottlenecks • Gigabit connectivity issues • 100 Mb/s in path • Defaulting to half duplex • Connection to Abilene 622 Mb/s • Multiple firewalls between transfer machine and LC edge • TCP stack configuration • Congestion and forwarding slowness

  17. Bottleneck Solutions • Gigabit problems to resolve: • RDC lab’s upstream connection configured initially as 100Mb/s half duplex, because of… • Bad cable (manifested as erratic/asymmetric transfers) • RDC Lab firewall appliance not up to task • Appliance spec’d for self-contained development lab environment • Replaced with Linux-based firewall/router on commodity x86 hardware • Firewall throughput increased from 20Mbps to 800Mbps • Required same TCP stack tuning as transfer endpoint • TCP stack tuning of endpoints: Emilio @ SDSC – Andy @ LC http://www.psc.edu/networking/projects/tcptune/

  18. LC Data Transfer: Network Environment

  19. Results after Optimization Iperf 2.02: LC  SDSC 388 Mbits/s

  20. Current Network Test Environment @ SDSC • Topology • TOOLS: BWCTL NDT THRULAY OWAMP • FUTURE NETWORK IMPLEMENTATION: Network Monitoring Test System outside SDSC Proactive resolutions Network Issues Periodical Automatic Network Tests

  21. Transfer Tools: Background • Project involved sending 6TB+, in 300-500GB packages, to SDSC, with rough goal of 1TB/day (thus 100Mb/s or better) for simplicity of transfer management • 600GB of scanned images: archival masters and smaller 'derivatives' • Web archives, produced by Heritrix web crawler (Internet Archive): 100MB containers • Secondary goal: prepare for LC to be able to change roles, and become receiver of transfers of similar content packages from other partners on Internet2 • NDIIPP partners • NDNP awardees • Internet Archive • more...

  22. Transfer Tool Selection;or, “why can’t we just use FTP?” • LC introduced to the problems of high-latency/bandwidth transfer • and relieved to learn that the scientific community faced and solved those problems years ago • SDSC recommended GridFTP/Globus Toolkit to best use bandwidth • LC wanted to learn about other common high-speed transfer tools as well • e.g. BBFTP, BBCP, FDT

  23. Transfer Tool Setup • SDSC helped with Globus setup on LC side, in a collision of two worldviews: • SDSC: connections not firewalled, but require GSI certificate-based authentication • LC: triply-firewalled/filtered lab • challenging to get all necessary ports open for all tools of interest to be able to participate as grid peer • SDSC needed to work within GSI authentication framework • Open question: will cultural institutions be able to “get on the grid”?

  24. Transfer Results • Achieved 160-280Mb/s transfer rates with GridFTP, and RFT atop GridFTP (providing resumption on failure). • Best single data rate transfer was with roughly an 8MB TCP window, and 2 parallel streams, yielding about 90 Mbps. Multiple GridFTP sessions led to 200-280Mbps • Some packages had many small files - individual JPEGs • packaging in TAR, or using tool that packages, improved throughput • Integrity checking via checksumming with MD5 • three hours per package • Can be parallelized across CPUs, but I/O can be limiting factor

  25. Other Transfer Tools • FDT (CERN/Stanford) still in active development, unlike other tools; seems well suited for integration into larger suites; Java-based • LC testing FDT transfers with other partners • BBFTP & BBCP • of same vintage as GridFTP • might be more useful if not using GSI and lower setup costs desired • …but GridFTP can now run in SSH-authenticated (or unauthenticated?) environment; not tested

  26. Next Steps for Preservation Partners • Repeating transfer process, in other direction: LC receiving content from NDIIPP and other Internet2-connected partners • Partners span spectrum of expertise and capability: some partners include transfer tool builders, while others are librarians who know their content but aren't IT/network-savvy; must deal with both • Need to establish approaches for simple network tuning • Transfer tool selection challenges • For a one-time transfer, what setup cost is acceptable? • GridFTP/Globus offers strong performance, but at high setup cost • Producing “decision tree” to help transfer partners make informed choices

  27. Thank you Andy Boyko - Library of Congress - aboy@loc.gov Jane Mandelbaum - Library of Congress - jman@loc.gov Robert McDonald - SDSC - mcdonald@sdsc.edu David Minor - SDSC - minor@sdsc.edu Emilio Valente - SDSC - evalente@sdsc.edu

More Related